Raj Patel

AI is scaling fastest where leaders simplify the stack, unify the platform, and treat data & infrastructure as a compounding advantage. • 97% say cloud platforms are critical to scaling AI • 93% are prioritizing energy efficiency to scale sustainably • 38% have unified their platform to simplify their stack, and they’re already seeing the upside If you’re making 2026 infrastructure decisions, this report shows where momentum is building, where friction remains, and how the leaders are pulling ahead: bit.ly/4qDuRC7

104

1 Comment

Cloud cost uncertainty isn’t a fluke; it’s by design. Cloud providers profit from complexity: 💸 Granular billing 💸 Shifting discount programs 💸 Opaque pricing for AI workloads 💸 Exploding SaaS and shadow IT The result? Engineering and FinOps leaders are left scrambling to explain unpredictable cloud bills while innovation can’t slow down. Traditional dashboards and static budgets won’t save you. You need adaptive cloud cost management that embeds cost awareness into engineering workflows, detects anomalies early, and drives shared accountability. We break it all down in our latest post, including 5 steps to start leading through the chaos. Read it here: https://lnkd.in/ge4jadxr #FinOps #CloudCost #EngineeringLeadership #CloudComputing #AI #SaaS #Ternary

35

1 Comment

🎯 Running inference workloads on Kubernetes? I just wrote a deep dive on Gateway API Inference Extension and how it works. The Gateway API Inference Extension introduces two new Custom Resource Definitions (CRDs): InferenceModel and InferencePool. By using these two new resources, along with the endpoint selector, your L7 routing infrastructure becomes an “Inference Gateway” enabling you to self-host GenAI/LLMs with a “model-as-a-service” mindset. https://lnkd.in/g_eFqwUd #ai #llm #genai #inference

57

2 Comments

Kubernetes cost optimization is more than just monitoring - it’s a strategy. This article compares Kubecost and OpenCost, helping you choose the right fit for your enterprise cloud setup. A must-read for DevOps, FinOps, and platform teams 🔗 Read here: https://lnkd.in/ddKzbkTe #Kubernetes #FinOps #DevOps #CloudCost #PlatformEngineering #PlatformEngineers

5

People often ask me about the differences in architecture between Amazon Dynamo (the 2007 SOSP paper), DynamoDB (the AWS serverless NoSQL database), and Aurora DSQL (the AWS serverless SQL databases). I captured the key points on durability, high availability, programming model, and trade-offs in my blog: https://lnkd.in/gkMG_tVD TL;DR: Despite the shared name, Dynamo and DynamoDB have little in common architecturally, and make very different trade-off decisions.

632

8 Comments

I love the way Ozgun puts it: Ubicloud is an open-source alternative to AWS, at a fraction of the cost How do they dare take on these cloud giants? Özgun shares the backgrounds of the founding team (Daniel and Umur) and what it really takes to go head to head with big tech   More videos from e2vc Summit!

64

1 Comment

Demand response participation is increasing as a result of increased DER adoption due to resiliency needs and long interconnection queues. The lost revenue due to not being powered is a bigger driver than capturing demand response revenue. I don't think there's an argument that demand response works. I do think there's an argument to be made that it's difficult to capture value from demand response alone. For that reason, I've argued that demand response products will have to be features of larger platforms. https://lnkd.in/gueF2QnK

12

2 Comments

As I'm leaving AWS re:Invent in Vegas, I’m reminded of a counterintuitive truth: competing with the public clouds is often really good business VCs love to ask, “Why wouldn’t the incumbent do this?” But in markets this large (and evolving this quickly) that’s rarely the right question. The real questions are: Who can assemble the best team, get to market fastest and with the most focus? And where are there opportunities to partner with the biggest players rather than compete? We’ve seen this pattern repeat across the ecosystem: -People assumed Vercel and AWS were destined to compete. Today, they’re strategic partners with CEO Guillermo Rauch in the keynote -Snowflake vs. Amazon Redshift. Now, AWS is one of Snowflake’s strongest distribution channels -In Datadog’s early pitches, the concern was, “If it’s just monitoring AWS, why not use CloudWatch?” Fast forward to now, and Datadog is now $50B+ public company We’re going to see the same dynamics play out with OpenAI and Anthropic. These are becoming foundational platforms. But that doesn’t preclude incredible companies from being built on, next to, and even around them. Some will look competitive. Many will end up as partners. And the biggest outcomes often start in those gray zones

217

26 Comments

Memory Centric Infrastructure - AI has fundamentally changed data-center economics. Memory—not compute—is now the primary constraint driving cost, utilization, and scalability. Yet most infrastructure remains CPU-centric, forcing enterprises to buy entire servers just to get more memory, stranding capital in underutilized systems and exposing businesses to volatile RAM and HBM markets. At TORmem, we are delivering a practical shift to memory-centric infrastructure: memory becomes a shared, rack-scale resource, and servers become clients. The result is higher GPU utilization, longer server lifecycles, and materially better control over infrastructure spend. Memory-centric architecture is no longer a research concept—it is becoming a business necessity.

27

Another AWS re:Invent update for you: what happened in the infrastructure keynote yesterday? This keynote is often, to me, the most interesting: they share stories of the incredible design and engineering choices that it takes to build a successful global hyperscaler cloud. This year, of course, the lens was the growing demand for AI services. AWS' custom ARM hardware, Graviton, gets a new launch: M9g EC2 instances, available in preview, sport new Graviton 5 chips. This doubles the number of cores vs Graviton 4, and provides increased L3 cache size, to improve performance by reducing cache misses. We got an official Dave Brown definition of Serverless: it is "the absence of server management". We learned that Amazon Bedrock was originally built in a "traditional web service" type of architecture, with load balancers at the front and servers at the back. The learnings from that approach drove Project Mantle, which offers latency tiering, and fair scheduling, to reduce the impact of noisy neighbours on your own workloads. Mantle uses a request journal, so that failed requests can be tried again, making Bedrock more reliable. AI workloads need access to vector databases. AWS have invested in making vector search available across the platform. S3's Vector indexes offer great price/performance, and we learned how they can still achieve speedy access to these indexes even though they're stored on disc, not in memory. AWS' new Trainium 3 chip processes 5x the number of tokens per MW than Trainium 2. These packages have on-chip support for profiling workloads without impacting performance. AWS are making Neuron Explorer (a performance analysis tool) available, and as of Q1, you'll be able to use the features of Trainium natively from PyTorch, just by targeting 'neuron' instead of 'cuda' in that code, using NKI - the Neuron Kernel Interface, initially announced in September last year.

12

📢📢Big announcement from OpenAI pivoting from proprietary to a more "open" approach doing justice to its name :) 𝐒𝐭𝐫𝐚𝐭𝐞𝐠𝐲: 🎯This is a great pivot for OpenAI to diversify beyond Microsoft Azure and open it up to other cloud players including its own with Oracle. 🎯This is a bold go to market move by OpenAI and is now really open and nicely challenges open and other proprietary vendors for cloud and more specifically edge. 🎯This will accelerate adoption of OpenAI models for research as well as commercial use under Apache 2.0 license. ✨ OpenAI dropped their first open-weight reasoning model models since GPT-2, dubbed GPT-OSS-120B and GPT-OSS-20B a chain-of-thought reasoning model. ✨ Open AI has trained these using a mix of reinforcement learning and techniques informed by OpenAI’s internal models, including o3 and other frontier systems. 𝐈𝐦𝐩𝐥𝐢𝐜𝐚𝐭𝐢𝐨𝐧𝐬: 🥊This shakes up the AI competitive landscape confirming "Open" is king and Meta's and other DeepSeek AI and Alibaba Cloud Qwen approach. ✨The gpt-oss-120B model to run within 80GB of memory, 𝐰𝐡𝐢𝐥𝐞 𝐠𝐩𝐭-𝐨𝐬𝐬-𝟐𝟎𝐛 𝐨𝐧𝐥𝐲 𝐫𝐞𝐪𝐮𝐢𝐫𝐞𝐬 𝟏𝟔𝐆𝐁. This is a big moment for bringing OpenAI to on-device and an inflection point for developers and enterprises to adopt and build proprietary AI experiences on it to run natively. 🤝Great to see partners such as Ollama, LMSTUDIO, vLLM, Hugging Face, Fireworks AI, Azure Foundry, AWS Sagemaker supporting and working with NVIDIA, Qualcomm and AMD to democratize on-device AI for these OpenAI models. ✨Qualcomm: By integrating Ollama’s lightweight, open-source LLM servicing framework with powerful Snapdragon platforms, developers and enterprises can run the gpt-oss-20b directly on devices with Snapdragon compute platforms and also run web search and several other features default out of the box ✨AMD: OpenAI GPT-OSS 20B and 120B language models for local AI inferencing on AMD products like the Ryzen™ AI processors and Radeon™ graphics are ready with day 0 support through LM Studio ✨NVIDIA: he easiest way to test these models on RTX AI PCs, on GPUs with at least 24GB of VRAM, is using the new Ollama app fully optimized for RTX. 𝐓𝐂𝐎: 💲From cost perspective, enterprises and developers looking to host the model themselves, fine tune and build proprietary experiences on top of it without per-use API fees which could be a minor initial upfront cost factor for many startups, researchers. However, they need to invest in their own infrastructure, talent and resources to host and develop in-house. So depending upon the enterprise AI strategy, there will be different TCO calculations. 💲Great move for enterprises with high-volume usage or mission critical needs vs low usage, where AI-as-a-Service will benefit better. This also now offers higher bargaining power for enterprise against other AI vendors and even AI-as-a-Service models. #OpenAI #GenAI #AIModels #Meta #NVIDIA #Qualcomm #AMD

308

22 Comments

We recently announced the GA of Oracle Database@AWS. Customers can now run Oracle Exadata Database Service and Oracle Autonomous Database on dedicated infrastructure on OCI within the AWS U.S. East (N. Virginia) and U.S. West (Oregon) Regions. In addition, to meet growing customer demand, we plan to make the service available in an additional 20 AWS Regions around the world. InfoWorld has the details: https://lnkd.in/gswnvM9z

109

8 Comments

Modern applications are no longer just code. They are complex systems spanning services, infrastructure, networks, storage, Kubernetes, and AI workloads. Legacy APM tools cannot look deep below the surface, leaving teams chasing symptoms instead of identifying the real problem. That is why I am proud to announce Virtana Application Observability! It automatically traces performance issues across the entire stack, from application code to infrastructure and AI workloads, surfacing evidence-backed root cause so teams and AI agents can act quickly and confidently. This is how enterprises move from fragmented observability to full system-level insight, faster triage, and more reliable operations. Learn more about how Application Observability can transform your enterprise operations: https://lnkd.in/gMkzfvpP Read Virtana's Press Release on Virtana Application Observability: https://lnkd.in/gK_qnAQP

13

Asked yesterday by a colleague: “Why did you folks just release IX-API v2.7.1? Weren’t you working full steam ahead on version 3?” That’s a perfectly fair question! 🧐 ➡️ IX-API — the open standard for Interconnection Services — continues to evolve over time. We need to support new use cases, enhance existing solutions, and fix the occasional tricky bug that slips through QA. With version 1 of the standard already scheduled for deprecation on July 1st, 2026, but version 3 still under development, the latest, newest supported version remains 👉👉👉 v2.7. Regardless of the resources it requires, we’ll continue to support and improve the v2.x line for a while. Even once v3 is ready, migration will take effort from clients — and not everyone can make that transition overnight! It’s a bit like what has also happened this week with Windows 10: even though Windows 11 has been out for some time, not everyone can migrate right away or at all. Microsoft has therefore introduced an Extended Security Updates program to continue supporting Windows 10 for another year — a very sensible decision for the good of the entire internet! 🏅 ✅ But I digress. In short: if you’re using IX-API v2, you’re in good shape — we, DE-CIX, AMS-IX, London Internet Exchange (LINX) have got you covered! ✅ More details about the IX-API standard here: https://ix-api.net #IXAPI #NetworkAutomation #NaaS #InterconnectionAsAService

12

2 Comments